Florida’s Orlando Police Department (OPD) and the city of Orlando are planning to launch a second round of testing of Amazon’s Rekognition facial recognition software, the parties announced Monday.
The announcement came despite a firestorm of protests earlier this year when the first pilot was conducted over several months. The American Civil Liberties Union (ACLU) protested the use of facial recognition technology as a law enforcement tool, while Amazon’s shareholders called for the company to forego selling its Rekognition technology to government agencies.
Despite the protests, Orlando Police Chief John Mina, Deputy Police Chief Mark Canty, Orlando Chief Administrative Officer Byron Brooks, and Orlando CIO Rosa Akhtarkhavari sent a memo to the Orlando mayor and commissioners on July 6, asking to continue the Amazon facial recognition pilot program. The group noted that while the initial pilot provided valuable information, more testing was needed before they could make a recommendation.
“We don’t know if the technology will prove to be successful, efficient and cost effective model that would enhance our public safety efforts. That is why we are doing a pilot, to internally test — for free — the technology, capabilities and interface with our existing systems and the technology we have in place,” Cassandra Lafser, the mayor’s press secretary, told Government Technology.
Under Amazon’s facial recognition program, the customer uploads two data sets. One data set is made up of images, such as a mugshots database, and the other is an input photo of the potential suspect. Rekognition then takes the input photo and scans the database for a potential match.
As part of second pilot, Orlando and the police department announced it will work with Amazon Web Services to refine their statement of work to include these four parameters:
If pilot is successful, the memo notes the Orlando Police Department’s legal team and the city attorney’s office would draft proposed policy and procedures for the city council to review. An Amazon spokesperson said the company requires its customers to abide by the law when using its services, and if it learns services are being abused, it will suspend the account.
Orlando is not the only government agency testing or using Amazon’s Rekognition facial recognition software. The Sheriff’s Office in Washington County, Ore., currently has Rekognition deployed.
“Amazon Rekognition has greatly increased the ability of our law enforcement officers to act quickly and decisively. We were able to index more than 300,000 photo records within 1-2 days, and the identification time of suspects went from 2-3 days down to minutes. Within one week of going live, the solution built on Rekognition identified a suspect for a cold case that led to an arrest through due process,” Chris Adzima, senior Information Systems Analyst for the Washington County Sheriff’s Office, said in a statement posted on Amazon’s site.
Meanwhile, the South Carolina Law Enforcement Division (SLED), Santa Barbara County, Calif., Sheriff’s Office, and Los Angeles County Sheriff’s Department use DataWorks Plus, which also develops facial recognition software, according to its customer list.
“I estimate that less than 5 percent of all law enforcement agencies in the United States use facial recognition, but five years from now it maybe closer to 10 percent,” Todd Pastorini, DataWorks Plus’ executive vice president and general manager, said.
He added that law enforcement agencies across the nation have embraced facial recognition in certain regional pockets but not others.
“The Northwest has been slow to deploy facial recognition and the same with Texas. Colorado, New Mexico and Montana also have nothing. But in the Northeast, and the states of Michigan and Pennsylvania, they have a statewide deployment of facial recognition,” Pastorini said, noting that Arizona also allows for the statewide deployment of facial recognition.
In statewide deployments, all law enforcement agencies in the state can access a central state database of images that can be checked against a specific photo of a potential suspect that is submitted. Although California has approved shared use of a statewide database for facial recognition, Pastorini said California has not “turned it on” for law enforcement agencies to access.
“In California, facial recognition is used on a county-by-county basis, as a result,” he said.
According to Pastorini, Los Angeles County recently conducted a 12-month test with Riverside County to access each other’s facial recognition repositories, and other nearby counties are considering joining in.
“As facial recognition technology improves, it gets used more frequently. And the more it is used, the more other agencies will hear about it,” he noted, pointing to its growth potential. “Some, however, will never understand it, or embrace its use, while others live in fear of Big Brother.”
And, of course, lack of funding is a reason Pastorini said he often hears when potential clients say they cannot explore facial recognition technology at this time.
For law enforcement agencies that have signed on, a number of them wanted mobile capabilities back in the 2009 to 2010 period, when the technology was starting to find a home with state and local law enforcement agencies, Pastorini said. Officers wanted to take a photo of someone off the street with a smartphone and run the individual through the facial recognition software for potential matches of suspects. Now, however, mobile facial recognition technologies are common, he noted.
One key takeaway for law enforcement agencies considering the use of facial recognition technology is that it is just one of many investigative tools that detectives could use and does not provide positive identification of a suspect or unidentified party, say officials with Pennsylvania’s Justice Network (JNET) and Oregon’s Washington County Sheriff’s office. Both agencies provide access to facial recognition to their respective law enforcement members.
“If facial recognition picks a potential candidate, you still have to do the investigative work. It only provides an additional clue and is a beginning point for an investigation,” said Eric Webb, director of Business and Service Management for Pennsylvania’s Public Safety Delivery Center.
The state’s Public Safety Delivery Center oversees JNET, which provides online access to all the booking photos held by various law enforcement agencies in Pennsylvania through its JNET Facial Recognition System (JFRS).
The technology is not used for mass surveillance or real-time surveillance, but rather is put to use when a crime has been committed and clear photos or video are available from the crime scene, said both Deputy Jeff Talbot, public information officer with the Washington County Sheriff’s office, and Webb.
The photos may come from an assault victim as the assailant flees after throwing a punch, or from store video camera as a robber approaches a clerk at the counter of a convenience store, both noted.
“We try to strike a balance between honoring civil liberties and fighting crime,” Talbot explained, adding that store camera footage and a photo taken of a detained suspect who is believed to be providing false information are the most common cases where the technology is used..
Most of the complaints the ACLU and privacy groups have expressed around facial recognition tend to center on a desire to have laws and policies in place that would guard against such acts as using the technology to conduct mass surveillance on citizens and real-time surveillance. Both Webb and Talbot say their agencies are able to avoid such concerns because of the internal policies, legislation and laws their state and local governments enacted prior to deploying the tools.
“The biggest issues we wanted to work out was developing standard operating procedures of when and how to use the technology,” Talbot recalled. Washington County’s Sheriff’s office has been using facial recognition technology for approximately 18 months, when it kicked off a pilot test that later evolved into full deployment.
JNET, which started using facial recognition approximately 10 years ago, requires law enforcement members to attend in-person classroom training on its facial recognition policies and laws and the technology it offers before they can access that investigative tool feature.
“You cannot do this training online,” Webb says. “You need classroom training to get a well-educated and skilled group of users who can do facial recognition. It’s a mistake to think you can do one online class and then roll this out to a bunch of people.”
Dawn Kawamoto is a former staff writer for Government Technology.